|At the ConferenceExhibitsTransportationLodgingDiningNightlife|
SC Conference - Activity Details
Megajobs: How to Run One Million JobsPrimary Session Leader:
Marlon Pierce (Indiana University)
Secondary Session Leaders:
As large systems surpass 200K CPU cores and as applications increase in complexity, more scientists need to run thousands to millions of closely related jobs that are associated with individual projects. Scientists seek convenient means to specify and manage many jobs, arranging inputs, aggregating outputs, identifying successful and failed jobs and repairing failures. System administrators seek methods to process extraordinary numbers of jobs for multiple users without overwhelming queuing systems or disrupting fair-share usage policies. Under development are a new generation of queuing and scheduling systems and multi-level schedulers for use with existing queuing and scheduling systems, schedulers designed to handle millions of jobs. This Birds-of-feather session provides a venue for the exchange of information about processing large numbers of jobs. Short presentations of an invited sample of projects will be followed by discussion.